library(tidyverse)
library(plotly)
library(sf) #for analyzing spatial data
library(rgdal) #note that `rgdal` requires another package `sp`; for importing zips shapefile, transforming CRS, and transforming coordinates into a SpatialPointsDataFrame
library(leaflet) #for interactive maps
library(tigris) #geojoin
knitr::opts_chunk$set(
fig.width = 10,
fig.asp = .2,
out.width = "90%",
dpi = 300,
warning = FALSE,
message = FALSE
)We first defined our dataframe crime_df_map.
#read crime_df data
crime_df = read_csv("data/crime_df.csv") %>%
janitor::clean_names() %>%
rename(case_number = x1)
#define crime_df_map for our map
crime_df_map = drop_na(crime_df, latitude)Get a first look of our dataframe by making a simple map draft. Drag the slider below to see spatial distribution of crime incidents in different time period during a day.
crime_plot =
crime_df %>%
filter(!is.na(borough)) %>%
plot_ly(
lat = ~latitude,
lon = ~longitude,
frame = ~hour_of_day,
type = "scattermapbox",
mode = "markers",
alpha = 0.2,
color = ~borough) %>%
layout(
title = "Spatial distribution of crime incidents",
mapbox = list(
style = 'carto-positron',
zoom = 9,
center = list(lon = -73.9, lat = 40.7)),
legend = list(title=list(text='<b> Borough </b>'))
) %>%
animation_slider(
currentvalue = list(prefix = "Hour of day: ", font = list(color="black"))
)
crime_plotWe want to draw a crime rate map to examine the crime rate in each zip code region. Crime rate for a zip code region is defined as number of crime incidents occurred in the zip code region during 2016-01-01 and 2017-12-31, divided by population size in the corresponding zip code region.
We first imported zips shapefile and transform CRS.
zips = readOGR("data/cb_2015_us_zcta510_500k/cb_2015_us_zcta510_500k.shp", verbose=FALSE)
zips = spTransform(zips, CRS("+proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0"))Then we matched each crime incident to a zip code region by utilizing
their coordinates variables longitude and
latitude. We included a zip variable to
crime_df_map data frame, representing corresponding zip
codes of each crime incident.
#extract only lon and lat
crime_lat_long = select(crime_df_map, longitude, latitude)
#transform coordinates into a SpatialPointsDataFrame
crime_spdf = SpatialPointsDataFrame(coords = crime_lat_long, data = crime_lat_long, proj4string = CRS("+proj=longlat +datum=WGS84 +ellps=WGS84 +towgs84=0,0,0"))
#subset only the zipcodes in which points are found
crime_zips = zips[crime_spdf, ]
#redefine crime_df_map to include zipcodes
crime_df_map = cbind(crime_df_map, over(crime_spdf, crime_zips[,"ZCTA5CE10"]))
crime_df_map = rename(crime_df_map, zip = ZCTA5CE10)300 crime incidents can’t be matched to a zip code, so we drop them.
crime_df_map =
crime_df_map %>%
drop_na(zip)In order to make meaningful data analysis, we want zip codes to
represent regions with similar population. However, some zip code
regions are very small, among which there are even some buildings that
have their own zip codes. To deal with the challenges of zip codes, we
convert zip codes to modified zip codes to account for discrepancies in
population size. Modified zip codes represent regions with similar
population size. In order to make the conversion, we use a conversion
table ZCTA-to-MODZCTA.csv obtained from here.
We first read the conversion table
ZCTA-to-MODZCTA.csv.
zcta_conv = read_csv("data/ZCTA-to-MODZCTA.csv")
zcta_conv$ZCTA = as.character(zcta_conv$ZCTA)
zcta_conv$MODZCTA = as.character(zcta_conv$MODZCTA)We matched zip in crime_df_map with
corresponding modified zip code, represented by a new variable
mod_zip.
crime_df_map = crime_df_map %>%
left_join(rename(zcta_conv, zip = ZCTA), by = "zip") %>%
rename(mod_zip = MODZCTA)4 crime incidents can’t be matched to a modified zip code and leave
NA values in the mod_zip column. This is because the 4
crime incidents happened in zip code 10550 or 10704, which are not in
New York City. Therefore, we dropped the 4 observations.
crime_df_map = drop_na(crime_df_map, mod_zip)In order to make an interactive map based on modified zip codes, we
use modzcta shapefile obtained from here.
Note this shapefile is based on modified zip codes instead of standard
zip codes.
modzcta = st_read("data/MODZCTA_2010/MODZCTA_2010.shp", quiet=TRUE) %>%
janitor::clean_names()We included shapefile information into crime_df_map.
crime_df_geo =
geo_join(modzcta, crime_df_map, "modzcta", "mod_zip", how = "inner") %>%
rename(mod_zip = modzcta)We can then obtain number of crimes in each zip region, sorting in descending order.
number_of_crimes =
crime_df_geo %>%
group_by(mod_zip) %>%
summarize(number_of_crimes = n()) %>%
arrange(desc(number_of_crimes))In order to obtain crime rate, we got population of each zip code region from here.
pop_zip = read_csv("data/csvData.csv") %>%
janitor::clean_names()
pop_zip$zip = as.character(pop_zip$zip)
#filter out regions in New York City
pop_zip = pop_zip %>%
filter(county %in% c("New York", "Kings", "Queens", "Bronx", "Richmond"))The dataset pop_zip lacks population data in zip code
region 10065, so we manually added it in (data
source here).
pop_zip[nrow(pop_zip) + 1,] = list("10065","
New York City", "New York", 30339)We then converted zip code to modified zip code, and got the population in each modified zip code region.
pop_mod_zip = pop_zip %>%
left_join(rename(zcta_conv, zip = ZCTA), by = "zip") %>%
rename(mod_zip = MODZCTA) %>%
group_by(mod_zip) %>%
summarize(population = sum(population))We added population of each modified zip code region into
number_of_crimes, and calculated crime rate in each
modified zip code region. Crime rates were sorted in descending
order.
crime_rate = left_join(number_of_crimes, pop_mod_zip, by = "mod_zip")
crime_rate$crime_rate = crime_rate$number_of_crimes / crime_rate$population * 100
crime_rate =
crime_rate %>%
select(mod_zip, crime_rate) %>%
group_by(crime_rate) %>%
arrange(desc(crime_rate))#label
labels = sprintf(
"<strong>%s</strong><br/>Crime rate is %g%<br/>during the period from 2016-01-01 to 2017-12-31.", crime_rate$mod_zip, crime_rate$crime_rate) %>%
lapply(htmltools::HTML)
#color palette
pal = colorBin(palette = "OrRd", 9, domain = crime_rate$crime_rate)
map_interactive = crime_rate %>%
st_transform(crs = "+init=epsg:4326") %>%
leaflet() %>%
addProviderTiles(provider = "CartoDB.Positron") %>%
addPolygons(label = labels,
stroke = FALSE,
smoothFactor = 0.5,
opacity = 1,
fillOpacity = 0.7,
fillColor = ~ pal(crime_rate),
highlightOptions = highlightOptions(weight = 5,
fillOpacity = 1,
color = "black",
opacity = 1,
bringToFront = TRUE)) %>%
addLegend("bottomright",
pal = pal,
values = ~ crime_rate,
title = "Crime cases per 100 residents<br/>during the period from<br/>2016-01-01 to 2017-12-31",
opacity = 0.7) %>%
addTiles("Interactive map of crime rate")
#saveWidget(map_interactive, "nyc_crime_rate_map.html")
map_interactiveWe cut date into 4 weeks interval, and want to visualize
crime counts and crime rate in each modified zip code region for every 4
weeks since 2016-01-01 by using R Shiny.
crime_month = crime_df_map %>% #use crime_df_map instead of crime_df_geo here because crime_df_geo is too large
group_by(mod_zip, date) %>%
summarize(crime_counts = n())
crime_month$date = cut(crime_month$date, "28 days")
crime_month = crime_month %>%
group_by(mod_zip, date) %>%
summarize(crime_counts = sum(crime_counts)) %>%
rename(month_following = date) %>%
filter(month_following != "2017-12-29") #observations for 2017-12-29 removed for too few data
#add back mod_zip, crime_rate, and geometry
crime_month = left_join(crime_month, pop_mod_zip, by = "mod_zip") %>%
select(mod_zip, population, month_following, crime_counts)
crime_month$crime_rate = crime_month$crime_counts / crime_month$population
crime_month = crime_month %>%
select(month_following, mod_zip, population, crime_counts, crime_rate) %>%
arrange(month_following)
crime_month = geo_join(modzcta, crime_month, "modzcta", "mod_zip", how = "inner")
crime_month$month_following = as.Date(crime_month$month_following, format = "%Y-%m-%d")saveRDS(crime_month, "r_shiny_map/crime_month.RDS") #save for R ShinyThe Shiny app can be accessed here.